Ideologies are the black holes of the human thought space. Their pull moves adherents past an event horizon that makes it impossible to consider ideas outside of the ideological frame.
@BroadbandNow your website no longer lets users select the kind of service(dsl/fiber/etc) for frontier communications on the availability map and as a result it is now very inaccurate
Claude 3.7 Sonnet is…
- a 10x dev
- with incredible taste
- that’s smarter than you
- but infinitely patient
- and delightful to work with
Literally the perfect teammate.
I honestly can’t believe we have a day-to-day coding model that’s this good - it’s unbelievable.
@olipdsanders@SuryaGanguli How do those links change anything with what me, Surya or Vivek said? Those articles were provided as examples to back up Surya's statements which I never disagreed with. Vivek didn't say that all govt spending had a low roi, only most of it.
@elonmusk It's not financially wise for most of the population to have kids in this current economic environment. Collapsing birth rates is a symptom of the hoarding of wealth by the 1%. Rebuild the middle class and you'll get your birth rates back up.
@BenHolfeld When ai is trained on those tests, it's going to score high on them. Ai is currently surprisingly clever, at times.. But when push comes to shove, it often fails in many areas that humans still do better. It is arguably smarter than most humans already, at least in some ways.
@Lucidicus22 @ylecun@docmilanfar I would like to say more, but the character limit makes it difficult to really dive into this discussion as deeply as I would like to.
@Lucidicus22 @ylecun@docmilanfar Filtered selection isn't compression though. We perceive as much as is necessary to intelligently navigate reality in ways that increased chances of survival. Eyes do process info before sending it down the optic nerve, but I don't think it's a matter of bandwidth on the "wire"
@brycedriesenga@ylecun@docmilanfar And our eyes likely are suboptimal. They definitely aren't very good or accurate.. I'm not suggesting that eyes are meticulously designed by divine engineers. I just want to know where he gets these numbers from because he uses numbers like this a lot in references to vision.
@brycedriesenga@ylecun@docmilanfar I think he was literally suggesting that our retinas need to compress data in order for it to fit through the optic nerve, which seems dubious to me. If it was just a loose analogy, fine. But he was asserting "facts" to support his premise that vertebrate eyes are suboptimal.
@ylecun@docmilanfar According to this, there's a reason for that despite seeming counter intuitive. Also, the photoreceptors are themselves neurons. I'm curious where you got the compression data, or why that would even be necessary to "squeeze" data through the optic nerve
https://t.co/qIk6qYDMBp
@shadbush@ylecun @DrHughHarvey I hope your data doesn't end up in ai models. They need quality data. By all means, choose another platform. Or better yet, just stop posting your thoughts anywhere. It's your safest bet to ensure your precious words are never used to train ai... 🙄
@roninhahn@teknium@ID_AA_Carmack @fuglypirate @ylecun I was pointing out that llms don't currently follow the same reasoning process as we do when they perform math stuff. I was comparing them to humans, not calculators. Though one chatbot I play with is great about "showing its work" when doing math - which I consider impressive.
@teknium@ID_AA_Carmack @fuglypirate @ylecun Very true.. Once we get ai to reliably use external tools like modern calculators along with agentic reasoning processes... Things are going to start getting very interesting..
@teknium@ID_AA_Carmack @fuglypirate @ylecun I consider this an important distinction... Models capable of doing math generally need to be shown tons of examples of math. However humans do math very differently.. We're taught a few rules, and can then deduce the answer by applying those rules. I don't think llms can do that
@teknium@ID_AA_Carmack @fuglypirate @ylecun I guess what John is saying is that animals are proof that vast amounts of knowledge/data ≠ "AGI".. As we understand AGI in the biological sense.